Current computing data center solutions typically include a number of servers having devices, such as memory, storage, processors, etc, physically located on multiple racks. These servers are generally configured in an identical manner so that software can run on any of the servers. However, this can be expensive in both money and power consumption. Further, not all software requires all of these devices. An alternative approach is to move the devices off the server and place them on the network as a shared resource for any server to access. However, the problem with this approach is that Ethernet networks, which are typically used in computing data centers, having very high latency that impedes performance.